#metabrainz

we only use docker for the test instance

(and mirrors)

great, thanks

mayhem: https://www.irccloud.com/pastebin/5RX0FjlT/

looks good, committing

👍

hi!

there are many algs out there to calculate BPM of tracks and they are all flawed, one way or another.

what works for one musical style, doesn't work for another.

so, our question is: can a neural network do a better job?

In my search too, I wasn't able to find an algorithm being used widely for this task. I think I will be needing some data to experiment and explore neural networks for this use case.Would it be possible to get some sample data at this point of time?

sure.

for starters, use some of your own music and convert it to wav files and run it through the alg. and if you get some promising results, I'll build a larger dataset

thanks, I will start experimenting and reach out soon with some results

ok

Hi all, I'm working on integrating MusicBrainz into a Letterboxd for music app. Noticed its search results tend to be broader than what Spotify's API offers.

Considering using Spotify for initial searches and mapping those to MusicBrainz entities. Thoughts?

Alternatively, thinking about customizing the SOLR engine on a MusicBrainz server for tighter search results. Is this feasible?

Would appreciate any insights or advice on these approaches or other ideas!

hi!

when you say broad search results, what do you mean? do you have an example?

sure. let me find an example

https://musicbrainz.org/search?query=alright&am... If i search for this on on Spotify, the first result is "Alright" by Kendrick Lamar which is what I would expect.

I know that's biased, but i'm wanting something that would default towards more popular results

ok, yeah, we've gotten that complaint before and it makes sense.

and yeah, the answer is to add popularity to the data, which should solve the problem.

do you have the MBID for that recording to hand?

b5e1e12f-8832-4ba4-b772-de15e2db2009 ?

yep

ok, let me try a few things.

awesome. much appreciated

hmmm. that track is not in our recording popularity data.

for a 2015 release, it should be. odd.

lucifer: do you know when the last time was that we calculated recording popularity? or is that being done periodically?

zima_blue: because what could be done is to take all the search results and then match them against recording popularity and then sort based on that. that ought to move that track near the top.

but we're not really ready to release that data -- and the given track isn't in that dataset in the first place, so we first need to see what is going on with that dataset. lucifer will know when he's around.

cool. thanks for the context. i'll keep an eye out for a response

@mayhem I also have a doubt if the goal is to find out the highest bpm value throughout the music or the one that remains for majority of the music?

the latter.

cool, thanks

Just for additional context regarding my earlier query https://musicbrainz.org/search?query=nevermind&... I would expect the Nirvana release to be closer to the top

mayhem: we do it weekly, its broken currently because of the recording mbid i fixed earlier today. i'll initiate a new dump and request the data again tomorrow once the dump is imported to spark.

diptanil: hi! the metabrainz notification system is quite a difficult project idea. i would suggest to get familiar with at least one of the projects, maybe ListenBrainz because that's where we'll start using it immediately, first.

as for the chatlogger project, we currently use https://github.com/metabrainz/brainzbot-core/ for logging, its an IRC bot. but we are planning to eventually be matrix first so we would like the chat logger to be also using matrix APIs. matrix has a sdk for bots which you can use to build it: https://github.com/turt2live/matrix-bot-sdk

lucifer: could the "brokenness" of the similar recordings by caused by the problem you fixed? because zima_blue's example track isn't in the DB, despite being from a popular album.

mayhem: similarity data hasn't been generated in a while, but popularity yes.

I swear my dyslexia is getting worse. sigh.

any idea why a popular track might be missing from the popular recordings data?

zima_blue: fwiw if i understand correctly letterboxd for music would not be much different from what ListenBrainz is.

we batch recordings into 5-10k while inserting data in db, if any single row is problematic entire batch fails.

so that would be my first guess.

ok, cool. then ping me when the data set is run and we'll have another look.

will do

also, you need to check if the recording is a canonical recording.

I did. it was.

ah cool

i just grepped the error logs with the mbid you had posted, its there.

it was part of a batch that failed to insert.

good good. I'll follow up with zima when they return

so most probably should be fixed the next time we generate this data.

let me do a couple of PR merges and then i'll do a release followed by creating new dumps

yes it does.

cool, i'll go ahead and merge then.

hang on.

let me re-write the sentence now that I understand the intent

great, thanks!

"Deleted listens present a tricky problem in this setup, since they are not included in the incremental dumps. To get a fully accurate list of listens, with deleted listens removed, you'll need to re-import a full dump." ??

sounds much better yes!

ok, good.

[listenbrainz-server] 14amCap1712 merged pull request #2777 (03master…master): LB-1499: Improved the documentation of incremental dumps https://github.com/metabrainz/listenbrainz-serv...

[listenbrainz-server] 14amCap1712 merged pull request #2785 (03master…patch-2): LB-1499: Update incremental dumps documentation https://github.com/metabrainz/listenbrainz-serv...

lucifer: Hey! I was trying to figure out, are we still using this endpoint anywhere since BrainzPlayer is now a react component? https://github.com/metabrainz/listenbrainz-serv...

<11YoMo12> Something changed on LB that there's missing popular tracks by artists? https://listenbrainz.org/artist/4d5abad8-d57b-4... https://listenbrainz.org/artist/cd2a8c6d-6c11-4...

lucifer: Thanks for the help. If I understand ListenBrainz correctly, I believe iCheckMusic has some overlap but there are also differences. iCheckMusic focuses more on reviewing/reflecting

There’s also an element of organizing: favoriting things, marking you want to listen to something later, lists, etc

ansh: not sure, best to ask monkey.

zima_blue: ah, nice to know.

YoMo12: yes should be fixed soon

aerozol: https://listenbrainz.org/explore/fresh-releases/

entire listen history, minimum 10 listens of artist threshold

lucifer: holy shit

game changer!!

A whole stack of albums I have to go listen to now/didn’t know were out!

I’ll ask Discord and #musicbrainz for feedback too

mayhem: I think it’s an excellent idea for you to ping Glenn if you’re keen! Since we follow each other you can also PM him from any of the MB accounts

I haven’t directly asked him to contribute because I feel he might be looking for paid work right now, so that feels weird. But that’s a discussion you may be able to have!

p.s. he may also be spoilt for data haha, after having direct access to Spotify for so long, but we can promise other benefits (like not scrapping a bunch of work on a whim)

ok, I'll see about doing that tomorrow.

lucifer: I think we broke the week/month/3 months filter for Fresh Releases :)

i like the changes, much better

I posted the changes to the socials for feedback, always good to drive traffic to check out Fresh Releases anyway

from @MetadataIsMyDrug on Discord: My thoughts: Too much for me. But for others it wasn’t enough before, and too much is better than not enough if those are the two options. But my tastes have shifted a number of times and I have very little interest in a lot of what I see now, and some days have up to 14 releases now. 😮